The Research Of Front-end Processing Technology Based On The Speaker-independent Speech Recognition

Recently,With the rapid development of the artificial intelligence,The technology of speech recognition has been gradually entered a stage of practical application from the research.It is highly and potentially valuable,but how to improve the recognition rate of speech recognition system at present is still one of the hot spots.The process of the speech recognition and principle are introduced systemically and the improved algorithm is proposed on the key points of the speech recognition.The processes of speech recognition include roughly: the voice activity detection,the feature extraction,linguistic model training and recognition algorithm.Firstly,the partial key techniques in this speech recognition is researched in this thesis,including preprocessing of voice signal,endpoint detection and feature extraction.Then,the algorithm of voice activity detection based on Empirical Mode Decomposition(EMD)with improved wavelet energy entropy and weighted autocorrelation method for pitch detection based on wavelet packet transform are proposed for low Signal-to-Noise Ratio(SNR)condition,and the results of traditional algorithms are analyzed and compared.Secondly,Mel cepstrum coefficient is selected as the characteristic parameters in this thesis,and this paper deeply analyzes the MFCC parameters extraction process,and an improved MFCC extraction algorithm based on wavelet packet transform was proposed.The experiment results show that the new feature parameter makes the speech recognition system more robust.The average recognition accuracy in three kinds of SNR environments is increased respectively compared to the traditional feature parameters that LPCC and MFCC.Meanwhile,the whole system based on Hidden Markov Model(HMM)is established.The results show that the recognition rate of the system is improved by improved algorithm after comparing with the traditional methods of endpoint detection and feature extraction.Finally,a GUI of the speech recognition system is designed and the processing of real time speech recognition based on speech library is demonstrated.